I've managed to avoid Windows GDI for decades, and now I'm paying the price. The following C++ code does not result in the correct colors. It's just supposed to white-fill a window and then alpha blend a single bitmap. All of the alpha is per-pixel alpha in the bitmap.
// Create a memory DC
HDC hdcMem = CreateCompatibleDC(hdc);
HBITMAP hbmMem = CreateCompatibleBitmap(hdc, cxWnd, cyWnd);
SelectObject(hdcMem, hbmMem);
// White-fill the BG
FillRect(hdcMem, &rectWnd, static_cast<HBRUSH>(GetStockObject(WHITE_BRUSH)));
// Alpha blend the bitmap into the memory DC
HDC hdcSrc = CreateCompatibleDC(hdcMem);
HBITMAP hbmOld = static_cast<HBITMAP>(SelectObject(hdcSrc, hbm));
BLENDFUNCTION bfn = { AC_SRC_OVER, 0, 255, AC_SRC_ALPHA };
AlphaBlend(hdcMem, x, y, cxBmp, cyBmp, hdcSrc, 0, 0, cxBmp, cyBmp, bfn);
SelectObject(hdcSrc, hbmOld);
DeleteDC(hdcSrc);
// Blit the memory DC to the screen
BitBlt(hdc, 0, 0, cxWnd, cyWnd, hdcMem, 0, 0, SRCCOPY);
I've verified that all the variable values are correct and that the bitmap in hbm has the correct pixel colors and alphas.
The result shows many wrong colors, e.g., a blue pixel with low alpha should be pale blue but instead is dark red. Pixels with high alpha are much closer to what they should be.
Any help appreciated. TIA.
The answer, from hints in the comments: Each color channel in each pixel must be premultiplied by the alpha before blending.
get DIB bits
for each pixel
r = r * alpha / 255
g = g * alpha / 255
b = b * alpha / 255
set DIB bits
MS does not make this stuff easy to find. There's not a hint of it in the only alpha blending sample I could find:
https://learn.microsoft.com/en-us/windows/win32/gdi/alpha-blending-a-bitmap